Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The rapid increase in web-based machine learning is driven by rapid adoption for integration with frameworks such as TensorFlow.js, ONNX Runtime Web, and emerging WebGPU backends. This makes it more difficult to develop, less portable, and less scalable to deploy with this fragmentation. It is in this context of web applications that the present paper proposes an architecture-independent framework of standard Web Components as a structure for encouraging reusable, encapsulated, and interoperable combinations of ML. The proposed system introduces a layer of modular components, supported by runtime bindings in the form of adapters, lifecycle management, gradual model loading, and secure execution controls. Its architecture allows for separating user interface logic from ML runtime dependencies, and for flexible frameworks and deployment options, such as client-side, server-side, and hybrid inference. Experimental evaluation across image classification, text inference, and tabular prediction tasks demonstrates inference latency within acceptable bounds of direct runtime integrations under representative benchmark conditions, with measurably reduced integration complexity and improved portability across frontend ecosystems. Its results show that Web Components offer one of the best opportunities for abstraction layers usable across all web ecosystems, enabling machine learning in a transformable, safe, and efficient manner.